Markdown table in PDF gets incorrect renderings due to longtable usage by Pandoc

[cderv]

Discussed in #13063

---
format: pdf
keep-tex: true
---

| a | b |
|---|---|
| 1 | 2 |


| a | b |
|---|---|
| 1 | 2 |

: Caption {#tbl-test}

Counter gets incremented

Because longtable is used by the first table

\begin{document}


\begin{longtable}[]{@{}ll@{}}
\toprule\noalign{}
a & b \\
\midrule\noalign{}
\endhead
\bottomrule\noalign{}
\endlastfoot
1 & 2 \\
\end{longtable}

\begin{longtable}[]{@{}ll@{}}
\caption{Caption}\label{tbl-test}\tabularnewline
\toprule\noalign{}
a & b \\
\midrule\noalign{}
\endfirsthead
\toprule\noalign{}
a & b \\
\midrule\noalign{}
\endhead
\bottomrule\noalign{}
\endlastfoot
1 & 2 \\
\end{longtable}




\end{document} 

It seems we missed something in our tests - the documents are compiling, but the counter fix is not working.

This could also be related to a change in LaTeX package - more complex to know.

[cderv]

We do some replacement with longtable* but only for raw tex written in the document

quarto-cli/src/resources/filters/quarto-post/latex.lua

Lines 574 to 588 in 2f24776

	RawBlock = function(raw)
	if _quarto.format.isRawLatex(raw) then
	local longtable_match, _ = _quarto.modules.patterns.match_in_list_of_patterns(raw.text, _quarto.patterns.latexLongtableEnvPatterns)
	if longtable_match then
	local caption_match = _quarto.modules.patterns.match_in_list_of_patterns(raw.text, _quarto.patterns.latexCaptionPatterns)
	if not caption_match then
	-- We need to use the most generic pattern (last of the list) as we want to replace the environment and keep any options
	-- (e.g. `\begin{longtable}[c]{ll}` -> \begin{longtable*}[c]{ll} in flextable)
	local longtable_pattern = _quarto.patterns.latexLongtableEnvPatterns[#_quarto.patterns.latexLongtableEnvPatterns]
	raw.text = raw.text:gsub(_quarto.modules.patterns.combine_patterns(longtable_pattern), "\\begin{longtable*}%2\\end{longtable*}", 1)
	return raw
	end
	end
	end
	end,

Here, we have Table AST that is written by Pandoc using longtable

https://github.com/jgm/pandoc/blob/826f487818e6b36ec1e6feeafaa2e9449ce47695/src/Text/Pandoc/Readers/LaTeX/Table.hs#L30-L32

Maybe pandoc should have a way for us to say 'don't use longtable' or at least use unnumbered longtable* 🤔

For now, we can probably post process with other stuff:

quarto-cli/src/format/pdf/format-pdf.ts

Line 344 in 2f24776

function pdfLatexPostProcessor(

We already process caption it seems

quarto-cli/src/format/pdf/format-pdf.ts

Lines 982 to 1028 in 2f24776

	const longtableBottomCaptionProcessor = () => {
	let scanning = false;
	let capturing = false;
	let caption: string | undefined;
	return (line: string): string | undefined => {
	const isEndOfDocument = !!line.match(/^\\end{document}/);
	if (isEndOfDocument && caption) {
	return `${caption}\n${line}`;
	} else if (scanning) {
	// look for a caption line
	if (capturing) {
	caption = `${caption}\n${line}`;
	capturing = !line.match(/\\tabularnewline$/);
	return undefined;
	} else {
	if (
	line.match(/^\\caption.*?\\tabularnewline$/) ||
	line.match(/^\\caption{.*}\\\\$/)
	) {
	caption = line;
	return undefined;
	} else if (line.match(/^\\caption.*?/)) {
	caption = line;
	capturing = true;
	return undefined;
	} else if (line.match(/^\\endlastfoot/) && caption) {
	line = `\\tabularnewline\n${caption}\n${line}`;
	caption = undefined;
	return line;
	} else if (line.match(/^\\end{longtable}$/)) {
	scanning = false;
	if (caption) {
	line = caption + "\n" + line;
	caption = undefined;
	return line;
	}
	}
	}
	} else {
	scanning = !!line.match(/^\\begin{longtable}/);
	}
	return line;
	};
	};

[cscheid]

Maybe pandoc should have a way for us to say 'don't use longtable' or at least use unnumbered longtable* 🤔

They've merged our typst:no-figure PR, so maybe they'd be receptive to this in LaTeX.